diff --git a/GPU/Common/ShaderUniforms.cpp b/GPU/Common/ShaderUniforms.cpp
index e92448c7ed..20dd1e1b14 100644
--- a/GPU/Common/ShaderUniforms.cpp
+++ b/GPU/Common/ShaderUniforms.cpp
@@ -34,7 +34,7 @@ void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool flipView
 		Uint8x3ToInt4_Alpha(ub->colorTestMask, gstate.getColorTestMask(), gstate.getAlphaTestMask());
 	}
 	if (dirtyUniforms & DIRTY_FOGCOLOR) {
-		Uint8x3ToFloat4(ub->fogColor, gstate.fogcolor);
+		Uint8x3ToFloat4_Alpha(ub->fogColor, gstate.fogcolor, NAN);
 	}
 	if (dirtyUniforms & DIRTY_SHADERBLEND) {
 		Uint8x3ToFloat4(ub->blendFixA, gstate.getFixA());
diff --git a/GPU/Common/ShaderUniforms.h b/GPU/Common/ShaderUniforms.h
index 55f08f7658..7db09bd724 100644
--- a/GPU/Common/ShaderUniforms.h
+++ b/GPU/Common/ShaderUniforms.h
@@ -59,7 +59,7 @@ R"(  mat4 proj_mtx;
   int spline_count_v;
   int spline_type_u;
   int spline_type_v;
-  vec3 fogcolor;
+  vec3 fogcolor; float nanValue;
   vec3 texenv;
   ivec4 alphacolorref;
   ivec4 alphacolormask;
@@ -84,7 +84,7 @@ R"(  float4x4 u_proj;
   int u_spline_count_v;
   int u_spline_type_u;
   int u_spline_type_v;
-  float3 u_fogcolor;
+  float3 u_fogcolor; float nanValue;
   float3 u_texenv;
   uint4 u_alphacolorref;
   uint4 u_alphacolormask;
diff --git a/GPU/Directx9/VertexShaderGeneratorDX9.cpp b/GPU/Directx9/VertexShaderGeneratorDX9.cpp
index 01fd322a5e..4313123d73 100644
--- a/GPU/Directx9/VertexShaderGeneratorDX9.cpp
+++ b/GPU/Directx9/VertexShaderGeneratorDX9.cpp
@@ -37,6 +37,16 @@
 
 namespace DX9 {
 
+// The PSP does not have a proper triangle clipper, but it does have a guard band and can rasterize rather large
+// triangles that go outside the viewport. However, there are limits, and it will drop triangles that are very
+// large. Some games appear to draw broken geometry, probably game bugs that were never discovered because the PSP
+// would drop the geometry, including Parappa The Rapper in an obscure case and Outrun. Try to get rid of those
+// triangles by setting the W of one of the vertices to NaN if they are discovered.
+const bool guardBandCulling = true;
+// Not sure what a good value for this is, it should probably depend on the framebuffer size.
+// Let's be conservative.
+const float guardBand = 64.0f;
+
 static const char * const boneWeightAttrDecl[9] = {	
 	"#ERROR#",
 	"float  a_w1:TEXCOORD1;\n",
@@ -344,6 +354,7 @@ void GenerateVertexShaderHLSL(const ShaderID &id, char *buffer, ShaderLanguage l
 		}
 	}
 
+
 	WRITE(p, "VS_OUT main(VS_IN In) {\n");
 	WRITE(p, "  VS_OUT Out;\n");  
 	if (!useHWTransform) {
@@ -369,26 +380,32 @@ void GenerateVertexShaderHLSL(const ShaderID &id, char *buffer, ShaderLanguage l
 		}
 		if (lang == HLSL_D3D11 || lang == HLSL_D3D11_LEVEL9) {
 			if (gstate.isModeThrough()) {
-				WRITE(p, "  Out.gl_Position = mul(u_proj_through, float4(In.position.xyz, 1.0));\n");
+				WRITE(p, "  float4 outPos = mul(u_proj_through, float4(In.position.xyz, 1.0));\n");
 			} else {
 				if (gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
-					WRITE(p, "  Out.gl_Position = depthRoundZVP(mul(u_proj, float4(In.position.xyz, 1.0)));\n");
+					WRITE(p, "  float4 outPos = depthRoundZVP(mul(u_proj, float4(In.position.xyz, 1.0)));\n");
 				} else {
-					WRITE(p, "  Out.gl_Position = mul(u_proj, float4(In.position.xyz, 1.0));\n");
+					WRITE(p, "  float4 outPos = mul(u_proj, float4(In.position.xyz, 1.0));\n");
 				}
 			}
 		} else {
 			if (gstate.isModeThrough()) {
-				WRITE(p, "  Out.gl_Position = mul(float4(In.position.xyz, 1.0), u_proj_through);\n");
+				WRITE(p, "  float4 outPos = mul(float4(In.position.xyz, 1.0), u_proj_through);\n");
 			} else {
 				if (gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
-					WRITE(p, "  Out.gl_Position = depthRoundZVP(mul(float4(In.position.xyz, 1.0), u_proj));\n");
+					WRITE(p, "  float4 outPos = depthRoundZVP(mul(float4(In.position.xyz, 1.0), u_proj));\n");
 				} else {
-					WRITE(p, "  Out.gl_Position = mul(float4(In.position.xyz, 1.0), u_proj);\n");
+					WRITE(p, "  float4 outPos = mul(float4(In.position.xyz, 1.0), u_proj);\n");
 				}
 			}
 		}
-	}  else {
+		if (lang != HLSL_DX9 && guardBandCulling) {
+			// Guard band culling
+			WRITE(p, "  float2 projPos = outPos.xy / outPos.w;\n");
+			WRITE(p, "  if (abs(projPos.x) > %f || abs(projPos.y) > %f) outPos.w = nanValue;\n", guardBand, guardBand);
+		}
+		WRITE(p, "  Out.gl_Position = outPos;\n");
+	} else {
 		// Step 1: World Transform / Skinning
 		if (!enableBones) {
 			// Hardware tessellation
@@ -574,18 +591,23 @@ void GenerateVertexShaderHLSL(const ShaderID &id, char *buffer, ShaderLanguage l
 		if (lang == HLSL_D3D11 || lang == HLSL_D3D11_LEVEL9) {
 			// Final view and projection transforms.
 			if (gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
-				WRITE(p, "  Out.gl_Position = depthRoundZVP(mul(u_proj, viewPos));\n");
+				WRITE(p, "  float4 outPos = depthRoundZVP(mul(u_proj, viewPos));\n");
 			} else {
-				WRITE(p, "  Out.gl_Position = mul(u_proj, viewPos);\n");
+				WRITE(p, "  float4 outPos = mul(u_proj, viewPos);\n");
 			}
 		} else {
 			// Final view and projection transforms.
 			if (gstate_c.Supports(GPU_ROUND_DEPTH_TO_16BIT)) {
-				WRITE(p, "  Out.gl_Position = depthRoundZVP(mul(viewPos, u_proj));\n");
+				WRITE(p, "  float4 outPos = depthRoundZVP(mul(viewPos, u_proj));\n");
 			} else {
-				WRITE(p, "  Out.gl_Position = mul(viewPos, u_proj);\n");
+				WRITE(p, "  float4 outPos = mul(viewPos, u_proj);\n");
 			}
 		}
+		if (lang != HLSL_DX9 && guardBandCulling) {
+			WRITE(p, "  float2 projPos = outPos.xy / outPos.w;\n");
+			WRITE(p, "  if (abs(projPos.x) > %f || abs(projPos.y) > %f) outPos.w = nanValue;\n", guardBand, guardBand);
+		}
+		WRITE(p, "  Out.gl_Position = outPos;\n");
 
 		// TODO: Declare variables for dots for shade mapping if needed.